14.13 spring 2020 recitation 10: projection bias and
TRANSCRIPT
Recitation 10: Projection Bias and Attribution Bias
Alex Olssen and Will Rafey
1
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Outline
1 Projection Bias
2 Attribution Bias
2
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Overview
Understand the influence of future vs. past states I Projection bias: mis–prediction of influence of future states I Attribution bias: mis–prediction of influence of past states
3
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Outline
1 Projection Bias
2 Attribution Bias
4
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Projection Bias
Image by Jeff on flickr. CC BY-NC-SA Image by harry_nl on flickr. CC BY-NC-SA
5
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Projection Bias: Model
True utility at time t is u(ct , st )
True utility depends on consumption ct and the state st at time t I State could be anything that affects utility from consumption, e.g., level of hunger or
addiction
At time t, predict future utility from consuming cτ in state sτ at time τ > t:
u(cτ , sτ ) = (1 − α)u(cτ , sτ ) + αu(cτ , st )
α ∈ [0, 1] is the degree of projection bias.
6
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Projection Bias: Example Problem
Maddie’s dog, Emma, suffers from severe projection bias. Emma’s true utility at hour t, u(ct , st )
u(ct = whine, st = hungry) = −3, u(ct = don0t whine, st = hungry) = −10, u(ct = whine, st = not hungry) = −5,
u(ct = don0t whine, st = not hungry) = 0
At time t, Emma predicts her future utility of whining at time τ > t to be
ut (cτ , sτ ) = (1 − α)u(cτ , sτ ) + αu(cτ , st ),
where α ∈ [0, 1].
7
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Projection Bias: α Parameter
What does α measure? I α measures the degree of projection bias
What does it mean for α to equal 0? I α = 0 means there is no projection bias (rational expectations)
What does it mean for α to equal 1? I α = 1 means there is full projection bias
8
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
wer y on
Projection Bias: Predicted Utility 3 Suppose α = 4 . Consider time period τ > t when Emma will be hungry sτ = hungry
How much utility will Emma expect to get from whining if she is hungry in period t?
ut (cτ , sτ ) = (1 − α)u(cτ , sτ ) + αu(cτ , st )
ut (whine, hungry) = (1 − α)u(whine, hungry) + αu(whine, hungry) 1 3
ut (whine, hungry) = −3( ) + −3( ) = −3 4 4
How much utility will Emma expect to get from whining if she is not hungry in period t?
ut (cτ , sτ ) = (1 − α)u(cτ , sτ ) + αu(cτ , st )
ut (whine, nothungry) = (1 − α)u(whine, hungry) + αu(whine, not hungry) 1 3
ut (whine, nothungry) = −3( ) + −5( ) = −4.5 4 4
Do the utilities differ? Why? Yes. If she is hungry at t, she correctly predict future utility. If she is not hungry at t, she
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020 incorrectly “projects” her lo utilit from whining to the future.
9
Outline
1 Projection Bias
2 Attribution Bias
10
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias
Definition I When judging the value of a good, people are overly influenced by the state in which they
previously consumed it
Examples I More likely to return to a restaurant first tried when hungry I More likely to negatively rate a movie seen while tired I Less likely to recommend a zoo to a friend if it rained during last visit
11
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias: Model
Predict utility of consuming ct while in state st , given prior consumption experience of ct was in state st−1(ct ):
u(ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
where γ ∈ [0, 1] is the degree of attribution bias.
Recall in the Projection bias model τ > t and
u(cτ , sτ ) = (1 − α)u(cτ , sτ ) + αu(cτ , st )
α ∈ [0, 1] is the degree of projection bias.
12
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias: Experimental Evidence (Haggag et al. 2019)
Basic structure I Randomly manipulate the thirst levels of participants I Have participants drink a new mixed drink I Elicit preference measures while in a later neutral state
Suppose participants who were thirsty say the like the drink better in the neutral state
Why can we take this as evidence of attribution bias? I.e., how do we know the state of prior consumption explains the difference?
13
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Experiment outline
(1)
(2)
(3)
(4)
(5)
Manipulation of thirst I Treatment group: drink 3 cups of water I Control group: drink 1/2 cup of water
Answer demographic questions, measure of current thirst (7-pt scale)
Stir together ingredients for new mixed drink: I 1 cup milk I 1/3 cup orange juice I 1 tablespoon sugar
Consume the mixed drink
Answer “how enjoyable was drinking the mixed drink?” (7-pt scale)
14
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Ingredient photo upload Ingredient photo upload
Attribution Bias in Economic Decision Making University of Chicago - Booth School of Business
15
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Summary of experiment
Experiment manipulated state (thirst) during new experience (drink)
Thirsty individuals like the drink better.
Since the two groups were randomized, we know that any difference must be due to the differential thirst across groups.
In the absence of attribution bias, people’s willingness to have the same drink again shouldn’t depend on thirst when first consuming it.
But people who were thirsty during their first experience are more willing to have the drink again.
16
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias Example Problem
Maddie’s true utility from exercising on day t, u(ct , st ) depends not only on the type of exercise ct (run or gym), but also on the state st (hot or cool):
u(ct = run, st = hot) = 5,
u(ct = gym, st = hot) = 6,
u(ct = run, st = cool) = 10,
u(ct = gym, st = cool) = 7.
17
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias: Predicted Utility Fall arrives and the weather in Cambridge today is st = cool Suppose Maddie most recently went to the gym when it was hot and ran when it was hot Will she choose to run or go to the gym?
I Maddie will choose the form of exercise that gives the higher predicted utility.
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (run, cool) = (1 − γ)u(run, cool) + γu(run, hot)
ut (run, cool) = 10(1 − γ) + 5γ
ut (run, cool) = 10 − 5γ
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (gym, cool) = (1 − γ)u(gym, cool) + γu(gym, hot)
ut (gym, cool) = 7(1 − γ) + 6γ
ut (gym, cool) = 7 − γ
I Maddie will run if 10 − 5γ > 7 − γ or if γ < 3/4 14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
18
Attribution Bias: Predicted Utility II Suppose Maddie most recently went to the gym when it was hot and ran when it was cool
Will she choose to run or go to the gym? I Maddie will choose the form of exercise that gives the higher predicted utility.
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (run, cool) = (1 − γ)u(run, cool) + γu(run, cool) = u(run, cool) = 10
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (gym, cool) = (1 − γ)u(gym, cool) + γu(gym, hot)
ut (gym, cool) = 7(1 − γ) + 6γ
ut (gym, cool) = 7 − γ
I Maddie will run if 10 > 7 − γ or if γ > −3. This is always true. I Intuition: Maddie last ran when it was cool (the best time to run). She last went to the gym
when it was hot (the worst time to go to the gym). Her attribution bias works in favor of running (which is her preferred choice when it is cool under her true utility function anyway)
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
19
Attribution Bias: Predicted Utility III
Suppose Maddie most recently went to the gym when it was cool and ran when it was cool
Will she choose to run or go to the gym? I Maddie will choose the form of exercise that gives the higher predicted utility.
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (run, cool) = (1 − γ)u(run, cool) + γu(run, cool) = u(run, cool) = 10
ut (ct , st ) = (1 − γ)u(ct , st ) + γu(ct , st−1(ct ))
ut (gym, cool) = (1 − γ)u(gym, cool) + γu(gym, cool) = u(gym, cool) = 7
I In this case, attribution bias has no effect because the current state is the same as the state for prior consumption
I Thus γ is irrelevant 20
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
Attribution Bias: Wrong Choices
In which of the three cases above and for which values of γ might we say Maddie makes the “wrong” choice?
When it is cool, best choice is to run
First case I Her prior run and gym visit were both when it was hot I Her predicted utility is biased because of attribution bias I If γ ≥ 3/4, then she will go to the gym (i.e., when she is close to full attribution bias)
Second case I Her predicted utility makes running even more preferable (because her past experiences
running were good and at the gym were bad) I She runs for any γ ∈ [0, 1]
Third case I Her predicted utility happens to have no bias (because the current state is the same as the
prior ones) I She runs for any γ ∈ [0, 1]
14.13 Recitation 10: Projection Bias and Attribution Bias April 23-24, 2020
21
MIT OpenCourseWare https://ocw.mit.edu/
14.13: Psychology and Economics Spring 2020
For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.
22